统计学习方法：KNN

KNN: 分类

也可用于回归
参考资料：KNN

"""
@File    : KNN.py
@Time    : 2020-7-30
"""

import pandas as pd
import numpy as np
import time
from collections import Counter

def loadData(fileName):
    #从文件中读取数据
    data=pd.read_csv(fileName,header=None)
    data=data.values #dataframe转化为ndarray
    y_label=data[:,0]  #第一列
    x_label=np.mat(data[:,1:])
    return x_label/255,y_label  #x_label归一化

def caculDistance(x1,x2): #计算两个点的距离 欧式距离 dis=sqrt(square（x1-x2）)  1x784 行向量有np.sum
    return np.sqrt(np.sum(np.square(x1-x2)))

def findCluster(x_train,y_train,x,k):  #x待分类数据 k最近k个点的k值
    distances=[]
    for xx in x_train:  #遍历：计算x与each样本的距离
        dis=caculDistance(xx,x)
        distances.append(dis)
    minK=np.argsort(distances)[:k]

    result=[]
    for i in range(k):  # 将对应分类结果添加到result中
        result.append(y_train[minK[i]])
    belonging=Counter(result).most_common(1)[0][0]  #只取元素
    return belonging

def tt(x_train,y_train,x_test,y_test,k):
    # 只测试50个
    acc_num,acc=0,0
    for i in range(50):   #range(len(x_test))
        cluster=findCluster(x_train,y_train,x_test[i],k)
        if cluster==y_test[i]: acc_num+=1
        print(f'find {i}th data cluster: cluster_pred={cluster},cluster={y_test[i]}')
        print('now_acc=',acc_num/(i+1))

if __name__=="__main__":
    start=time.time()
    x_train, y_train=loadData('Mnist/mnist_train.csv')
    x_test, y_test = loadData('Mnist/mnist_test.csv')
    tt(x_train,y_train,x_test,y_test,k=20)
    end=time.time()
    print('run time=',end-start)

—————————————————————————
新收获命令

# Counter对象：任意可哈希（hashable）元素构成的序列对象
Counter.most_common(k)  #Top k
out:[('eyes', 8), ('the', 5), ('look', 4)] #(元素,出现次数) k=3

#argsort()返回的是数组值从小到大的索引值
#argsort()[num]返回前0~k-1个索引值
x=np.array([1,4,3,-1,6,9])
x.argsort()
Out[3]: array([3, 0, 2, 1, 4, 5], dtype=int64)